Satlyt Raises $8M to Deploy AI Workloads Across Satellite Fleets

Satlyt, a Sunnyvale‑ and Nairobi‑based startup founded by former Google Cloud and SpaceX product manager Rama Afullo, has closed an $8 million seed round. The company does not build spacecraft; instead it creates a software platform that lets any satellite run AI models and share compute resources, much like VMware or Snowflake abstract infrastructure for data centers.

Afullo describes the vision as an "Android" for orbital data centers — an open, horizontally integrated ecosystem that works across vehicles from SpaceX, Google’s Project Suncatcher, and emerging players such as Starcloud and Cowboy Space Company. The platform has already flown on two demonstration missions, including a test of Google DeepMind’s Gemma model on a Momentus vehicle that reduced downlink volume for error logs by more than 60 %, translating into hundreds of thousands of dollars in annual savings per satellite.

This week Satlyt’s software launches aboard a TakeMe2Space satellite carrying payloads for NASA, space‑surveillance startup Stellerian, and TakeMe2Space itself. The mission will validate cloud‑computing protocols, on‑orbit image processing, and the ability to host third‑party workloads on a single spacecraft.

Why It Matters for GPU/AI Infrastructure

Orbital AI demands compact, radiation‑hard GPUs and efficient software orchestration — exactly the challenges AiGpu addresses with its cloud GPU fleet and hardware‑aware scheduling. Satlyt’s hardware‑agnostic layer could become a standard runtime for distributing inference and training jobs across a constellation, turning each satellite into a revenue‑generating compute node and expanding the addressable market for space‑qualified GPU services.